Software engineer used AI-assisted coding to build a weekend Home Assistant + Raspberry Pi audio monitor that correlates night noises with Garmin sleep data on a DAW-style timeline.
Key Takeaways
Two USB mics (indoor + outdoor), a Raspberry Pi rolling buffer, and a Home Assistant automation gate recording to only activate when the user is home and in bed.
The web app stitches three data sources (Garmin sleep stages, Home Assistant sensor events, Pi audio clips) onto a synced multi-track timeline; wake events are highlighted in red as the primary navigation signal.
AI did architecture and code generation, not sound classification; the author still listens manually, using the timeline to skip to moments worth reviewing.
Identified culprits: neighbor door slams, dish sounds, motorbikes, and trash trucks; fixes included IKEA acoustic panels, door/window sealing, and one household conversation.
Total build time was ~8 hours; author explicitly would not publish the code due to no formal review, but the home-lab network restriction limits exposure.
Hacker News Comment Review
Commenters noted Garmin is consistently rated among the worst smartwatches for accurate sleep-stage tracking in comparative tests, though wake-event detection is considered adequate for this use case.
CO2 levels visible in the screenshots drew attention as a potentially significant sleep quality factor independent of noise, with one commenter sharing that a cracked window kept bedroom CO2 below 600 PPM even near a motorway.
Several builders shared prior DIY equivalents using phone mics and Python amplitude-threshold scripts, confirming the core approach is sound but AI tooling compressed iteration time substantially.
Notable Comments
@nevi-me: CO2 in the screenshots looks unhealthy; bedroom ventilation may affect sleep quality beyond just acting as a noise source.
@phainopepla2: Raises that 3am waking is a known cortisol-spike pattern, and sleep-tracking anxiety can itself worsen sleep.
